Kelly Wen
Co-Founder, EdPilot
Most university AI policies were drafted in a hurry, by people who were primarily thinking about risk. A policy that starts with what you're trying to accomplish educationally produces something more useful.
Most university AI policies were drafted in a hurry, by people who were primarily thinking about risk. The questions driving the drafting process were: What can students abuse? What are we liable for? What do we need to prohibit?
Those are legitimate questions. They're not the right starting questions.
A policy that starts with "what are the risks we need to mitigate" produces a prohibition list. A policy that starts with "what are we trying to accomplish educationally, and how does AI affect our ability to accomplish it" produces something more useful — a governance framework that actually shapes the learning environment in productive ways.
What AI tools will the institution make available, and under what conditions?
If the institution isn't providing structured AI tools, students will use unstructured ones. The policy question isn't really "will students use AI" — it's "what AI environment will students be in." An institution that provides curriculum-grounded AI gives students something better than a prohibition gives them.
How will faculty exercise governance over AI in their courses?
Faculty authority over course content and academic standards should extend to AI. The policy should establish that faculty have the right to configure AI for their courses and provide the infrastructure to make that practically possible. A policy that says "faculty may set expectations about AI use" without giving faculty actual tools to shape AI behavior is largely symbolic.
What constitutes appropriate use in different assessment contexts?
The answer varies significantly across assignment types, disciplines, and learning objectives. A blanket policy that treats all AI use the same misses the important distinctions. The policy should establish a framework for thinking about this — not a single rule, but a principled way of distinguishing contexts where AI supports learning from contexts where it substitutes for it.
How will the institution learn and adapt?
AI capabilities are changing quickly. A policy written today will need revision within 18 months. Build in a review cadence. Establish who is responsible for tracking developments in AI capability and educational research on AI in learning. Make the policy a living document rather than a one-time exercise.
Treating this as an academic integrity policy. Academic integrity is one dimension of this. It's not the whole framework. A policy that's primarily about what students aren't allowed to do misses the affirmative question of what good AI-enabled learning looks like.
Writing for today's tools. If the policy specifies particular platforms or particular capabilities, it'll be outdated quickly. Write for principles and let implementation guidance handle specifics.
Excluding faculty from the drafting process. Policy written without faculty input tends to reflect administrative and legal concerns more than educational ones. Faculty governance structures should have a substantial voice in how AI policy is developed.
Treating compliance as the goal. A policy that students follow because they're afraid of getting caught produces different behavior than a policy that reflects genuine shared understanding of what good academic practice looks like in an AI-enabled environment. Aim for the latter.
A well-designed AI policy doesn't just mitigate risk. It creates conditions in which AI can be used productively — where students know what's expected, faculty have the tools to govern the AI in their courses, and the institution has visibility into what's actually happening in the learning environment.
That foundation makes the educational case for AI possible: structured tools that extend what faculty can teach, that support students outside of office hours, that give faculty feedback about where their courses are working. None of that happens without a governance framework that makes it safe to try.